<p>Lung nodule detection is necessary for lung cancer treatment, which is crucial for treating the patients. However, the current dataset comprises only a limited amount of lung Computed Tomography (CT) images, which shows potential imbalance between non-nodule and nodule samples. This disparity reduces the performance of neural networks and makes training more difficult. Early detection of lung tumor is crucial for identifying patients with a high chance of effective treatment, and this relies heavily on the precise recognition of malignant lung nodules in CT scans. Recently, the Deep Neural Network (DNN)techniques have been successfully employed to address various computer vision challenges, demonstrating their potential in this domain. Yet, low training values present a significant challenges in diagnosing the malignant nodules. The size and shape of a lump are crucial factors in determining malignancy in lung cancer. Therefore, it is important to address the limitations of conventional techniques by leveraging the deep learning strategies. The developed lung nodule classification framework contains three main stages: image collection, segmentation, and classification. Initially, a required CT images are gathered from publicly available sources. These images are processed through a segmentation module, whereas an effective segmentation is performed using a 3D Trans-DenseUnet++ (3D-TDUnet++) model. This segmentation process segregates the lung nodules from nearby tissues and eliminates the irrelevant background structures, which helps the developed model focus only on the region of interest. Also, it enhances the feature extraction process and reduces noise from images, which strengthens the classification accuracy and overall reliability of the developed method. After segmentation, the attained segmented images are further given into the classification phase by using an Adaptive DenseNet combined with a Long Short-Term Memory (LSTM) layer (ADNet-LSTM).This synergy of the classification-based deep learning model empowers an automated differentiation between malignant and benign lung nodules, which yields an accurate and efficient clinical decision-making process. It efficiently reduces the manual effort, improves the early detection process and also enhances the detection consistency. Additionally, the parameters of an ADNet-LSTM are optimized with the help of Intensified Fitness-based Red-Tailed Hawk Algorithm (IF-RTHA). Furthermore, extensive experimental validations are performed over the developed models with several performance metrics to ensure model’s reliability. The accuracy of the IF-RTH-ADNet-LSTM model is 94.98% higher than the existing works, such as RAN, Densenet, LSTM, and ADNet-DenseNet, as 90.21%, 92.38%, 91.68%, and 92.87% using the lung nodule dataset. Thus, it is revealed that the developed lung nodule classification framework performed well for evaluating the malignancy risk of lung nodules found on CT images and also it has the efficiency to provide better decisions for clinicians.</p>